Condensation-Concatenation Framework for Dynamic Graph Continual Learning

TL;DR

This paper introduces a novel framework called CCC for continual learning on dynamic graphs, which condenses historical data into semantic representations and selectively combines them with current data to mitigate forgetting caused by structural changes.

Contribution

The paper proposes a new condensation-concatenation framework that effectively preserves historical information and adapts to topological changes in dynamic graph continual learning.

Findings

CCC outperforms state-of-the-art methods on four real-world datasets.

The refined forgetting measure improves adaptation to structural updates.

The framework effectively mitigates catastrophic forgetting in dynamic graphs.

Abstract

Dynamic graphs are prevalent in real-world scenarios, where continuous structural changes induce catastrophic forgetting in graph neural networks (GNNs). While continual learning has been extended to dynamic graphs, existing methods overlook the effects of topological changes on existing nodes. To address it, we propose a novel framework for continual learning on dynamic graphs, named Condensation-Concatenation-based Continual Learning (CCC). Specifically, CCC first condenses historical graph snapshots into compact semantic representations while aiming to preserve the original label distribution and topological properties. Then it concatenates these historical embeddings with current graph representations selectively. Moreover, we refine the forgetting measure (FM) to better adapt to dynamic graph scenarios by quantifying the predictive performance degradation of existing nodes caused by structural updates. CCC demonstrates superior performance over state-of-the-art baselines across four real-world datasets in extensive experiments.